I'm suggesting a set of performance improvements to Key/Sha1 operations.
Basically it's a replacement of for loops with memcmp/memset/memcpy
And use Uint32*/Uint64* instead of Uint8* to perform calculations on several bytes at once. Internal Uint8[20] buffer declaration was changed to Uint32[5], but pointer to Uint8* is still returned via getData().
Also new test cases are added to keytest executable.

I used a valgrind function profiler in QtCreator to test performance